test(integration): align remaining release qualification checks (#49714)

* test(integration): align remaining release qualification checks

* test(integration): scope offline resolution to the restricted runner

.circleci/scripts/run_integration.sh

@@ -128,6 +128,8 @@ fi
 sudo iptables -L integration_only -n -v -x > "$results/egress-guard.txt"
 awk '$3 == "REJECT" && $1 > 0 { rejected=1 } END { exit !rejected }' "$results/egress-guard.txt"
 
+export UV_OFFLINE=true
+
 if [ "$mode" = parity ]; then
   export INTEGRATION_WORKERS=1
 else

tests/integration/observability/test_cache_hit_guardrail_metrics.py

@@ -253,6 +253,16 @@ def _count(samples: tuple, name: str, model_id: str) -> float:
     )
 
 
+def _deployment_states(rig: Rig) -> tuple[float, ...]:
+    return tuple(
+        sample.value
+        for sample in _metric_samples(rig.candidate, rig.model_name)
+        if sample.name == DEPLOYMENT_STATE
+        and sample.labels.get("model_id") == rig.deployment_id
+        and sample.labels.get("api_base") == rig.provider.url + "/v1"
+    )
+
+
 def _populated_failures(samples: tuple, rig: Rig, api_provider: str) -> float:
     return float(
         sum(
@@ -558,7 +568,8 @@ def test_cache_hit_post_call_guardrail_outage_keeps_deployment_labels(
         )
         assert reject.status_code >= 400, reject.text
         assert rig.provider.received.qsize() == 1, rig.provider.drain()
-        _expect_metrics(rig, 0, 0, pf_populated=1, pf_blank=0)
+        _expect_metrics(rig, 1, 0)
+        _assert_spend(_call_id(reject), rig)
 
 
 def test_two_identical_cache_hit_rejects_increment_populated_series(gateway: Gateway, tmp_path: Path) -> None:
@@ -635,6 +646,7 @@ def test_cache_hit_reject_deployment_metric_set_diff(gateway: Gateway, tmp_path:
         text: Final = "cache hit control e4 " + marker
         warm: Final = rig.candidate.request("POST", "/v1/chat/completions", _chat_body(rig.model_name, text, None))
         assert warm.status_code == 200, warm.text
+        eventually(lambda: _deployment_states(rig), lambda states: states == (0.0,))
         reject: Final = rig.candidate.request(
             "POST", "/v1/chat/completions", _chat_body(rig.model_name, text, rig.guardrail_name)
         )
@@ -642,11 +654,4 @@ def test_cache_hit_reject_deployment_metric_set_diff(gateway: Gateway, tmp_path:
         samples: Final = _expect_metrics(rig, 1, 0)
         blank: Final = tuple(sample for sample in samples if sample.labels.get("model_id") == "")
         assert blank == (), blank
-        states: Final = tuple(
-            sample.value
-            for sample in samples
-            if sample.name == DEPLOYMENT_STATE
-            and sample.labels.get("model_id") == rig.deployment_id
-            and sample.labels.get("api_base") == ""
-        )
-        assert states == (1.0,), states
+        eventually(lambda: _deployment_states(rig), lambda states: states == (1.0,))

tests/integration/pricing/test_price_data_reload_prices.py

@@ -284,7 +284,7 @@ def _charged_as_listed(proxy: Gateway, deployment: _Deployment, scale: float) ->
     assert _listed_rates(proxy, deployment.name) == _expected_listing(rates), deployment.name
     request_id, charged = _billed(proxy.client, proxy.key, deployment)
     expected: Final = _expected_cost(rates, deployment.usage)
-    assert charged == (None if rates is None else pytest.approx(expected, rel=1e-6)), deployment.name
+    assert charged == pytest.approx(expected, rel=1e-6), deployment.name
     return _Charge(request_id=request_id, expected=expected)
 
 

tests/integration/providers/test_decisions_wire.py

@@ -262,11 +262,28 @@ def _upstream_calls(gateway: Gateway, handle: ScenarioHandle) -> list[dict[str,
     return _calls_to(_observed_requests(gateway), handle)
 
 
-def _sdk_inference_call(gateway: Gateway, handle: ScenarioHandle, provider: _Provider) -> dict[str, JsonValue]:
+def _sdk_inference_call(
+    gateway: Gateway,
+    handle: ScenarioHandle,
+    provider: _Provider,
+    *,
+    metadata_authorization: str | None = None,
+) -> dict[str, JsonValue]:
     observed: Final = _upstream_calls(gateway, handle)
     inference_path: Final = f"/{handle.scenario_id}/v1/systemone"
     (call,) = tuple(request for request in observed if request["path"] == inference_path)
-    metadata_calls: Final = tuple(request for request in observed if request["path"] != inference_path)
+    environment_metadata: Final = {
+        "path": f"/{handle.scenario_id}/api/show",
+        "authorization": metadata_authorization,
+        "body": {"name": provider.body_model},
+        "method": "POST",
+        "api_key": "",
+    }
+    metadata_calls: Final = tuple(
+        request
+        for request in observed
+        if request["path"] != inference_path and (metadata_authorization is None or request != environment_metadata)
+    )
     assert metadata_calls in (
         (),
         (
@@ -556,7 +573,9 @@ def test_ollama_decisions_sdk_respects_environment_and_explicit_auth(
             api_key=api_key,
         )
         assert response.model_dump(mode="json") == _answer_body(provider)
-        call: Final = _sdk_inference_call(gateway, handle, provider)
+        call: Final = _sdk_inference_call(
+            gateway, handle, provider, metadata_authorization=authorization if environment_base else None
+        )
         assert (call["path"], call["authorization"], call["body"]) == (
             f"/{handle.scenario_id}/v1/systemone",
             authorization,